3 ms·

Well, there's no force that makes traffic "rise to fill roads"; there's externalities to having increased traffic capacity that may cause transportation needs t

by Cushman 14y ago

Well, there's no force that makes traffic "rise to fill roads"; there's externalities to having increased traffic capacity that may cause transportation needs to increase. We still have a lot of underutilized roads. The main difference in terms of capacity, I think, is that once we have majority- and fully-autonomous roads, the system will actually not allow the road to exceed capacity. Individual vehicles will respond to signs of lower throughput by getting off of the potentially congested road, while centralized signaling systems will help coordinate routes and departure times to optimize-- and due to the non-Newtonian nature of traffic, keeping throughput high actually increases road capacity. So if you get into your car at 9am and say, "Take me to work," it's not going to say, "Time with traffic: One hour," it's going to say, "We should depart in twenty-five minutes." Even better, if you are subscribed to a service like most Americans will be, you'll instead tell them the day or week or month before, "I need to be at work by 9am," and they'll say, "There will be a fifteen dollar capacity fee. Would you rather arrive at 8:30 or 10?" And let you and your work decide exactly how much of the economy's money it's worth for you to be there at exactly the same time as everyone else in the time zone.

bcoates 14y ago
That's way beyond driverless car, that's a global optimization problem. If it were possible, you wouldn't even need driverless cars to implement it. System-to-car communication and slightly more complicated street lights would be enough. For example, many cities already have lit signs that forbid turns from cross-streets onto oversubscribed roads, your nav system could allow you to pay the congestion fee and get authorization to disregard the sign and take the turn. Similarly, you could get out your smartphone and queue yourself to a destination and be told when you get to leave and take the fast route for free. As long as queuing at home got you to your destination faster than queuing in traffic, it would be the individual best strategy to do it and enough people would use it for it to work. The hard part is probably doing traffic scheduling at an acceptable latency, for a city like LA you'd need to model the behavior of millions of cars however many iterations your optimization algorithm needs every few seconds. These kind of problems are really computationally expensive, it may not be economical with current systems and algorithms.
- Cushman 14y ago
  > That's way beyond driverless car, that's a global optimization problem. If it were possible, you wouldn't even need driverless cars to implement it. System-to-car communication and slightly more complicated street lights would be enough. I disagree, I'd say scheduling is the smallest part of it. On most trips I take, there's a single major road that covers most of the distance, and I'm most likely to be delayed by traffic on that road slowing or halting completely. It's the moment-to-moment perverse behavior of other drivers when it comes to merging, changing lanes, accelerating and braking that causes that to happen, and having intelligent drivers on the same roads at the same times would make a much more significant impact. In other words, you're saying we could do global scheduling with traffic lights, which is true. I'm saying that with sufficiently competent drivers we won't need the lights, and I think that will cause a qualitative shift in the nature of traffic.

by Cushman 14y ago

bcoates 14y ago

That's way beyond driverless car, that's a global optimization problem. If it were possible, you wouldn't even need driverless cars to implement it. System-to-car communication and slightly more complicated street lights would be enough. For example, many cities already have lit signs that forbid turns from cross-streets onto oversubscribed roads, your nav system could allow you to pay the congestion fee and get authorization to disregard the sign and take the turn. Similarly, you could get out your smartphone and queue yourself to a destination and be told when you get to leave and take the fast route for free. As long as queuing at home got you to your destination faster than queuing in traffic, it would be the individual best strategy to do it and enough people would use it for it to work. The hard part is probably doing traffic scheduling at an acceptable latency, for a city like LA you'd need to model the behavior of millions of cars however many iterations your optimization algorithm needs every few seconds. These kind of problems are really computationally expensive, it may not be economical with current systems and algorithms.

Cushman 14y ago
> That's way beyond driverless car, that's a global optimization problem. If it were possible, you wouldn't even need driverless cars to implement it. System-to-car communication and slightly more complicated street lights would be enough. I disagree, I'd say scheduling is the smallest part of it. On most trips I take, there's a single major road that covers most of the distance, and I'm most likely to be delayed by traffic on that road slowing or halting completely. It's the moment-to-moment perverse behavior of other drivers when it comes to merging, changing lanes, accelerating and braking that causes that to happen, and having intelligent drivers on the same roads at the same times would make a much more significant impact. In other words, you're saying we could do global scheduling with traffic lights, which is true. I'm saying that with sufficiently competent drivers we won't need the lights, and I think that will cause a qualitative shift in the nature of traffic.